canadian hydrographic conference · naouraz brahim1,2 & sylvie daniel1 didier gueriot2, basel...

Naouraz BRAHIM1,2 & Sylvie DANIEL1

Didier GUERIOT2, Basel SOLAIMAN2

1: Centre de Recherche en GéomatiqueDépartement des sciences géomatiques

Université Laval2: Département Image et Traitement de l’Information

2: TELECOM Bretagne, Brest, France

3D reconstruction of underwater scenes

using image sequences from acoustic camera

Canadian Hydrographic conferenceJune 21- 23

Quebec

Contents

Introduction

3D reconstruction methodology

Feature point detection Edge detection Contour matching Multiscale analysis Salient point detection

Conclusion & Future work

Climate change is gradually affecting the underwater environment :– Temperature increases, chemistry modifications– Ocean circulation perturbation – etc.

Influencing:– Resources, population dynamics – Underwater structure stability

Monitoring and observation of underwater environment is a necessity to: describe, understand, model ... in order to protect the marine resources and structures and to prevent climate changes

Introduction

Introduction

Underwater observation is currently carried out using imagery systems:– Optical systems:

• Provides physical properties (color, reflection, geometry)

• Image quality depends on underwater conditions

• Limited range

– Acoustical systems• Image quality less affected by turbid water• Long range

Acoustical imaging: the most reliable observation mode in the underwater environment

KLEIN sidescan sonarSource: www.usinenouvelle.com

Acoustical Cameras

Introduction

Underwater structure 900 KHz 2D BlueView imaging sonarhttp://www.blueview.com/2d-Imaging-Sonar.html

Dual frequency IDentification SONarhttp://www.soundmetrics.com

Damaged concrete matsDIDSON

http://www.soundmetrics.comSalmon swimming

http://www.soundmetrics.com

DIDSON Diver with Complete Diver-Held System

DIDSON on RAUVER ROV

Bridge Pierhttp://www.soundmetrics.com

But..

Acoustical cameras provide 2-D image, they do not resolve the altitude of the observed scene this is a limitation when monitoring and to observing the underwater environment

Purpose of this work: designing and developing a method that enables 3D reconstruction of underwater scenes

Introduction


Multiple acquisition of the same underwater scene over an image sequence under different points of view

Exploitation of this information redundancy to reconstruct in 3D the observed scene

The developed methodology is inspired from stereovision techniques


3D reconstruction

3-Transformation and elevation estimation

2- Feature matching

1- Feature point detection

Image sequence

Global methodology

Feature point detection

General approach

1- Edge detection

2- Contour matching

3- Multiscale analysis

4- Salient point detection

Image filtered by Canny filter

Bridge sequence

Ship sequence

Original image

1- Edge detection Robust feature points clearly belong to object edges in the

collected images– Application of Canny filter to detect contours

Original image Image filtered by Canny filter

: Length of contour

: stands for location in image 1 and 2 : defines a small neighborhood around the location in image 2 : equal 1 if x is true, 0 if not

Hypothesis: Small camera movement between two successive images

2- Contour matching

Determining the percentage of contour points present in successive image frames within a small neighborhood

Purpose: Remove false contours and preserve object contours

frame i+1

frame i+1frame i

frame i

Ship sequence

2- Contour matching

Contour matching process

frame i+1

frame i+1frame i

frame i

Bridge sequence

2- Contour matching

Contour matching process

Scale 1

Scale 2

Scale k

Scale 3

Preserve only robust contours that will yield to robust corners and accurate camera pose estimation

Analyze corresponding contours through a multi-scale approach

At each scale:1. Gaussian filter2. Edge detection3. Contour matching4. Computation of the number

of corresponding contours in two successive frames

which scale allows robust contours extraction?

3- Multi scale analysis

Scale 1 (original contours)

Scale 2

Scale 4

Scale 6

Relevant contours from ship sequence

Corresponding contours from scale 1


Scale 1 (original contours) Scale 6

Scale 8Scale 2

Relevant contours from bridge sequence

Corresponding contours from scale 1


Contour salient points : points with high curvature

Corner Detector : Douglas-Peucker algorithm

Decrease the number of points in a curve, keeping only corner candidates points with highest curvatures


Frame i

Frame i+1

Ship sequence

Frame i+15

Example of corresponding points


Frame i

Frame i+1

Frame i+15

Bridge sequence


Example of corresponding points

Conclusion & future work

This research work proposes:

– A new methodology to

extract robust feature points

that enable 3D reconstruction

of underwater scenes

– This approach is based on

contour extraction, multi-scale

analysis and corner extraction

The proposed method demonstrates good performances and promising results

The next step will consist in devising the reconstruction methodology relying on these results:• Feature point matching• Computation of the camera movement• 3D reconstruction from corner points

Conclusion & future work

Questions ?

CFQCU : Conseil franco-québécois de coopération universitaire

Acknowledgement

To know more :

Naouraz Brahim: [email protected] Guériot : [email protected]

Sylvie Daniel : [email protected]

Telecom Bretagne

Université Laval

canadian hydrographic conference · naouraz brahim1,2 & sylvie daniel1 didier gueriot2, basel...

Documents